Application of Soft Computing Techniques in Optimizing Neural Network Training for Resource-Constrained Devices

The deployment of artificial intelligence on edge devices such as smartphones, IoT sensors, and wearable gadgets faces significant challenges due to limited computational resources, memory, and battery life. Traditional training methods like Stochastic Gradient Descent often result in high energy consumption and slow convergence. This paper presents a hybrid soft computing approach that combines Particle Swarm Optimization (PSO) with a Fuzzy Logic Controller to dynamically adjust learning rate and momentum during neural network training. Experiments conducted on simplified MNIST and CIFAR-10 datasets in a simulated resource-constrained environment demonstrate a 28% reduction in training time and 35% decrease in relative energy consumption while maintaining competitive accuracy. The proposed method offers an efficient and interpretable solution for adaptive learning on low-power devices, contributing to the emerging field of edge AI optimization.